https://github.com/valentinlibouton/weather_prediction
Prediction meteo par machine learning et deep learning (in progress)
https://github.com/valentinlibouton/weather_prediction
kaggle-dataset
Last synced: 4 months ago
JSON representation
Prediction meteo par machine learning et deep learning (in progress)
- Host: GitHub
- URL: https://github.com/valentinlibouton/weather_prediction
- Owner: ValentinLibouton
- License: mit
- Created: 2024-02-05T09:16:03.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2024-02-16T20:54:08.000Z (over 2 years ago)
- Last Synced: 2024-12-31T11:41:39.624Z (over 1 year ago)
- Topics: kaggle-dataset
- Language: Jupyter Notebook
- Homepage:
- Size: 9.39 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Models
1. ## `ML_SGDClassifier.joblib`
- Meilleurs paramètres trouvés :
{'classifier__alpha': 0.001, 'classifier__loss': 'log_loss', 'classifier__max_iter': 1000, 'classifier__penalty': 'l2'}
- Best score: 0.8429646485528453
- Rapport de classification :
| | precision | recall | f1-score | support |
|--------------|-----------|--------|----------|---------|
| False | 0.86 | 0.95 | 0.90 | 18828 |
| True | 0.72 | 0.48 | 0.58 | 5364 |
| accuracy | | | 0.84 | 24192 |
| macro avg | 0.79 | 0.71 | 0.74 | 24192 |
| weighted avg | 0.83 | 0.84 | 0.83 | 24192 |
- Précision du modèle sur l'ensemble des données de test : 0.8257275132275133
2. ## `best_model_in_deep_learning.h5` - une epoch
- Meilleurs paramètres trouvés :
{'model__activation': 'relu', 'model__dropout_rate': 0.3, 'model__kernel_regularizer': None, 'model__learning_rate': 0.001}
- Best score: 0.8442659139633178
| | precision | recall | f1-score | support |
|--------------|-----------|--------|----------|---------|
| False | 0.87 | 0.95 | 0.90 | 18828 |
| True | 0.72 | 0.48 | 0.58 | 5364 |
| accuracy | | | 0.84 | 24192 |
| macro avg | 0.79 | 0.72 | 0.74 | 24192 |
| weighted avg | 0.83 | 0.84 | 0.83 | 24192 |
- Précision sur les données de test : 0.8444940476190477
3. ## `best_model_in_deep_learning.h5` - 100 epochs
- Meilleurs paramètres trouvés :
{'model__activation': 'relu', 'model__dropout_rate': 0.3, 'model__kernel_regularizer': None, 'model__learning_rate': 0.001}
- Best score: 0.8448859333992005
| | precision | recall | f1-score | support |
|--------------|-----------|--------|----------|---------|
| False | 0.86 | 0.96 | 0.90 | 18828 |
| True | 0.74 | 0.45 | 0.56 | 5364 |
| accuracy | | | 0.84 | 24192 |
| macro avg | 0.80 | 0.70 | 0.73 | 24192 |
| weighted avg | 0.83 | 0.84 | 0.83 | 24192 |
- Précision sur les données de test : 0.843584656084656
4. ## `ML_SVClassifier.joblib`
- Meilleurs paramètres trouvés :
{'classifier': SVC(), 'classifier__C': 10, 'classifier__kernel': 'rbf'}
- Best score: 0.8494750705633111
| | precision | recall | f1-score | support |
|--------------|-----------|--------|----------|---------|
| False | 0.86 | 0.96 | 0.91 | 18828 |
| True | 0.77 | 0.45 | 0.57 | 5364 |
| accuracy | | | 0.85 | 24192 |
| macro avg | 0.81 | 0.71 | 0.74 | 24192 |
| weighted avg | 0.84 | 0.85 | 0.83 | 24192 |
- Précision sur les données de test : 0.836102843915344
5. ## `best_model_in_deep_learning_balanced.h5` - 100 epochs + balanced
- Meilleurs paramètres trouvés :
{'model__activation': 'relu', 'model__dropout_rate': 0.3, 'model__kernel_regularizer': None, 'model__learning_rate': 0.01}
- Best score: 0.7865973353385926
| | precision | recall | f1-score | support |
|--------------|-----------|--------|----------|--|
| False | 0.77 | 0.83 | 0.80 | 18958 |
| True | 0.81 | 0.75 | 0.78 | 18834 |
| accuracy | | | 0.79 | 37792 |
| macro avg | 0.79 | 0.79 | 0.79 | 37792 |
| weighted avg | 0.79 | 0.79 | 0.79 | 37792 |
- Précision sur les données de test : 0.7903259949195597
6. ## `ML_SGDClassifier_balanced.joblib` - balanced
- Meilleurs paramètres trouvés :
{'classifier__alpha': 0.001, 'classifier__loss': 'hinge', 'classifier__max_iter': 1000, 'classifier__penalty': 'l2'}
- Best score: 0.7781747080124555
| | precision | recall | f1-score | support |
|--------------|-----------|--------|----------|--|
| False | 0.77 | 0.79 | 0.78 | 18958 |
| True | 0.78 | 0.76 | 0.77 | 18834 |
| accuracy | | | 0.78 | 37792 |
| macro avg | 0.78 | 0.78 | 0.78 | 37792 |
| weighted avg | 0.78 | 0.78 | 0.78 | 37792 |
- Précision sur les données de test : 0.7492061812023709
7. ## `best_model_in_deep_learning_change_layers.h5`
- Meilleurs paramètres trouvés :
{'model__activation': 'relu', 'model__batch_size': 16, 'model__dropout_rate': 0.3, 'model__kernel_regularizer': None, 'model__learning_rate': 0.001}
- Best score: 0.7868949890136718
| | precision | recall | f1-score | support |
|--------------|-----------|--------|----------|--|
| False | 0.81 | 0.76 | 0.78 | 18958 |
| True | 0.77 | 0.82 | 0.79 | 18834 |
| accuracy | | | 0.79 | 37792 |
| macro avg | 0.79 | 0.79 | 0.79 | 37792 |
| weighted avg | 0.79 | 0.79 | 0.79 | 37792 |
- Précision sur les données de test : 0.7881297629127858
- Temps d'exécution: 22h42 pour 810 fits
```python
GridSearchCV(cv=5,
estimator=Pipeline(steps=[('scaler', StandardScaler()),
('model',
)]),
param_grid={'model__activation': ['relu', 'sigmoid', 'tanh'],
'model__batch_size': [16, 32, 64],
'model__dropout_rate': [0.3, 0.4, 0.5],
'model__kernel_regularizer': [None,
],
'model__learning_rate': [0.001, 0.01, 0.1]},
verbose=2)
```